AI search visibility in 2026: how to get cited by ChatGPT, Perplexity and Google AI Overviews

What changed, in numbers

Read those together and the picture is not "AI killed search". Google is still where the volume is, but Google now answers a fifth of queries itself and keeps most of the clicks. AI assistants are growing fast from a small base and send little traffic — but they increasingly shape what people believe before they ever search. Semrush's analysis found commercial-intent queries triggering AI Overviews grew 71% in six months, and its modelling suggests an AI-search visitor is worth about 4.4 times a traditional organic one, because they arrive pre-decided.

The shift also changed who gets seen. Similarweb's data shows ChatGPT's share of AI traffic fell from about 76% to 53% in a year while Gemini rose to roughly 27%. Optimising for "ChatGPT" is already too narrow; the target is the class of answer engines.

Citation is not ranking

The most important finding for anyone with an SEO habit is that the old signals do not transfer cleanly.

• Only 12% of URLs cited by ChatGPT, Gemini and Copilot rank in Google's top 10 for the query. Perplexity is closer to Google at 28.6%. (Ahrefs, August 2025)
• Google's own AI Overviews used to cite top-10 pages 76% of the time. By March 2026 that had dropped to 38%, as "query fan-out" lets the model pull from pages that rank for related sub-questions rather than the head term. (Ahrefs, 863,000 keywords)
• Across 680 million citations, Wikipedia accounts for 47.9% of ChatGPT's top-ten source share and Reddit for 46.7% of Perplexity's; only about 11% of domains are cited by both. (5WPR State of AI Citations 2026)
• YouTube alone accounts for about a fifth of AI Overview citations; Google-owned properties together are around 43%. (Ahrefs / 5WPR)

So the models look in different places than Google's ten blue links, they look at different pages of the same site, and each engine has its own favourite sources. That is why a page can rank #1 and never be cited, and why a page ranking #14 for the head term can be the one ChatGPT quotes.

What actually correlates with being cited

Ahrefs ran the largest correlation study on this — 75,000 brands, August 2025, updated in December. The result reorders the SEO playbook.

The interpretation is uncomfortable for link builders and liberating for everyone else: models cite what the web talks about. Being mentioned — in articles, forum threads, videos, podcasts, other people's comparisons — matters three times more than being linked. Schema markup, the tactic every "GEO guide" leads with, did nothing measurable in the one controlled test I could find. The Princeton/Georgia Tech GEO paper (the one that named the field) found content-level changes — adding statistics, citations and quotations — could lift generative-engine visibility by up to 40%, but that is about the page itself, not its markup.

The citation-readiness checklist

Here is what I now do to every page I want cited, in order of evidence behind it.

• Answer the question in the first 60 words, with the number in it. Models extract the direct answer. "SEnuke AI costs $79, $147 or $497 per month" gets quoted; "pricing is competitive" does not.
• Put original, sourced statistics on the page. The GEO paper's biggest lift came from adding statistics and citations. Cite the primary source with a date. Models — and the humans who write the mentions models learn from — prefer pages that show their work.
• Name entities consistently. Your brand, your product, the category, the founder — spelled the same way everywhere, on your site and off it. Ambiguity kills citation because the model cannot resolve who you are.
• Write for sub-questions, not just the head term. With top-10 pages now supplying only 38% of AI Overview citations, the pages that get pulled in are the ones that answer the specific follow-ups. An FAQ section that actually answers questions is worth more than it was.
• Get mentioned where the models look. Reddit and YouTube are not optional for Perplexity and Google respectively. One honest thread, one walkthrough video, mentioning your brand by name, does more than ten guest posts.
• Track citations, not just rankings. You cannot fix what you cannot see. Tools below.
• Keep the page dated and updated. Models weight recency for anything with a price or a date in it. A changelog on the page is a cheap freshness signal and an honesty signal for humans.

I used this list on my own guide to writing product reviews that rank and get cited, and on every page of this hub. Whether it works is measurable, which brings us to tooling.

What AI visibility tracking costs in 2026

Every major SEO vendor now sells this, almost always as a separate line item. Prices from each vendor's pricing page, 4 September 2026.

The economics are simple. A solo operator wanting to track AI citations is looking at $79–$199 a month for a dedicated tool. SEnuke AI's argument is that the same capability sits inside a $79 platform alongside the strategy layer. If its AI Citation module measures even a fraction of what Brand Radar does, the platform pays for itself on that line alone; if it is a token feature, the comparison collapses. That is the first thing I will test on launch day, and the review says so.

What not to do

• Do not block the AI crawlers reflexively. If GPTBot, PerplexityBot and Google-Extended cannot read you, you cannot be cited. Block training if you want; allow retrieval.
• Do not stuff pages with "what is X" headers hoping to be extracted. Models extract answers, not headers. One clear answer beats twenty signposts.
• Do not buy links to chase citation. Referring domains correlate at 0.22. The money is better spent on one good video.
• Do not measure success by AI referral traffic alone. It is 0.33% of B2B traffic and growing 303% a year. The influence is upstream of the click; track brand search volume alongside it.

Frequently asked questions